Академический Документы
Профессиональный Документы
Культура Документы
Tutorial
Gregory Piatetsky-Shapiro
KDnuggets
© 2006 KDnuggets
Outline
Introduction
Data Mining Tasks
Classification & Evaluation
Clustering
Application Examples
2
© 2006 KDnuggets
Trends leading to Data Flood
More data is generated:
Web, text, images …
Business transactions,
calls, ...
Scientific data:
astronomy, biology, etc
More data is captured:
Storage technology
faster and cheaper
DBMS can handle bigger
DB
3
© 2006 KDnuggets
Largest Databases in 2005
Winter Corp. 2005
Commercial Database
Survey:
1. Max Planck Inst. for
Meteorology , 222 TB
2. Yahoo ~ 100 TB (Largest
Data Warehouse)
www.wintercorp.com/VLDB/2005_TopTen_Survey/TopTenWinners_2005.asp
3. AT&T ~ 94 TB
4
© 2006 KDnuggets
Data Growth
5
© 2006 KDnuggets
Data Growth Rate
Twice as much information was created in
2002 as in 1999 (~30% growth rate)
Other growth rate estimates even higher
Very little data will ever be looked at by a
human
6
© 2006 KDnuggets
Knowledge Discovery
Definition
Knowledge Discovery in Data is the
non-trivial process of identifying
valid
novel
potentially useful
and ultimately understandable patterns in data.
from Advances in Knowledge Discovery and Data
Mining, Fayyad, Piatetsky-Shapiro, Smyth, and
Uthurusamy, (Chapter 1), AAAI/MIT Press 1996
7
© 2006 KDnuggets
Related Fields
Machine
Learning Visualization
Statistics Databases
8
© 2006 KDnuggets
Statistics, Machine Learning
and
Data Mining
Statistics:
more theory-based
more focused on testing hypotheses
Machine learning
more heuristic
focused on improving performance of a learning agent
also looks at real-time learning and robotics – areas not part of data
mining
Data Mining and Knowledge Discovery
integrates theory and heuristics
focus on the entire process of knowledge discovery, including data
cleaning, learning, and integration and visualization of results
Distinctions are fuzzy
9
© 2006 KDnuggets
Knowledge Discovery Process
flow, according to CRISP-DM
see
Monitoring www.crisp-dm.org
for more
information
Continuous
monitoring and
improvement is
an addition to CRISP
10
© 2006 KDnuggets
Many Names of Data
Mining
Data Fishing, Data Dredging: 1960-
used by statisticians (as bad name)
© 2006 KDnuggets
Some Definitions
Instance (also Item or Record):
an example, described by a number of attributes,
e.g. a day can be described by temperature,
humidity and cloud status
Attribute or Field
measuring aspects of the Instance, e.g. temperature
Class (Label)
grouping of instances, e.g. days good for playing
13
© 2006 KDnuggets
Major Data Mining Tasks
Classification: predicting an item class
Clustering: finding clusters in data
Associations: e.g. A & B & C occur frequently
Visualization: to facilitate human discovery
Summarization: describing a group
Deviation Detection: finding changes
Estimation: predicting a continuous value
Link Analysis: finding relationships
…
14
© 2006 KDnuggets
Classification
Learn a method for predicting the instance class
from pre-labeled (classified) instances
Many approaches:
Statistics,
Decision Trees,
Neural Networks,
...
15
© 2006 KDnuggets
Clustering
Find “natural” grouping of
instances given un-labeled data
16
© 2006 KDnuggets
Association Rules &
Frequent Itemsets
Transactions
Frequent Itemsets:
TID ProduceMilk, Bread (4)
Bread, Cereal (3)
1 MILK, BREAD,
Milk, Bread, CerealEGG
(2)
…
2 BREAD, SUGAR
3 BREAD, CEREAL
Rules:
Milk => Bread (66%)
4 MILK, BREAD, SUG
17
© 2006 KDnuggets
Visualization & Data Mining
Visualizing the data
to facilitate human
discovery
Presenting the
discovered results in
a visually "nice" way
18
© 2006 KDnuggets
Summarization
© 2006 KDnuggets
Classification
Learn a method for predicting the instance class
from pre-labeled (classified) instances
Many approaches:
Regression,
Decision Trees,
Bayesian,
Neural Networks,
...
Linear Regression
w0 + w1 x + w2 y >= 0
Regression computes
wi from data to
minimize squared
error to ‘fit’ the data
Not flexible enough
23
© 2006 KDnuggets
Regression for Classification
Any regression technique can be used for
classification
Training: perform a regression for each class, setting
the output to 1 for training instances that belong to
class, and 0 for those that don’t
Prediction: predict class corresponding to model with
largest output value (membership value)
24
© 2006 KDnuggets
Classification: Decision
Trees
2 5 X
25
© 2006 KDnuggets
DECISION TREE
An internal node is a test on an attribute.
A branch represents an outcome of the test, e.g.,
Color=red.
A leaf node represents a class label or class label
distribution.
At each node, one attribute is chosen to split training
examples into distinct classes as much as possible
A new instance is classified by following a matching
path to a leaf node.
26
© 2006 KDnuggets
Weather Data: Play or not
Play?
Outlook Temperature Humidity Windy Play?
sunny hot high false No
sunny hot high true No Note:
overcast hot high false Yes Outlook is the
rain mild high false Yes Forecast,
rain cool normal false Yes no relation to
rain cool normal true No Microsoft
overcast cool normal true Yes email program
sunny mild high false No
sunny cool normal false Yes
rain mild normal false Yes
sunny mild normal true Yes
overcast mild high true Yes
overcast hot normal false Yes
rain mild high true No
27
© 2006 KDnuggets
Example Tree for “Play?”
Outlook
sunny
overcast rain
Humidity Yes
Windy
No Yes No Yes
28
© 2006 KDnuggets
Classification: Neural Nets
29
© 2006 KDnuggets
Classification: other
approaches
Naïve Bayes
Rules
Support Vector Machines
Genetic Algorithms
…
See www.KDnuggets.com/software/
30
© 2006 KDnuggets
Evaluation
© 2006 KDnuggets
Evaluating which method works
the best for classification
No model is uniformly the best
Dimensions for Comparison
speed of training
speed of model application
noise tolerance
explanation ability
32
© 2006 KDnuggets
Comparison of Major
Classification Approaches
Train Run
time Time
34
© 2006 KDnuggets
Evaluation issues
Possible evaluation measures:
Classification Accuracy
Total cost/benefit – when different errors involve
different costs
Lift and ROC curves
Error in numeric predictions
35
© 2006 KDnuggets
Classifier error rate
Natural performance measure for
classification problems: error rate
Success: instance’s class is predicted correctly
Error: instance’s class is predicted incorrectly
Error rate: proportion of errors made over the
whole set of instances
Training set error rate: is way too
optimistic!
you can find patterns even in random data
36
© 2006 KDnuggets
Evaluation on “LARGE” data
37
© 2006 KDnuggets
Classification Step 1:
Split data into train and test
sets THE PAST
Results Known
+
+ Training set
-
-
+
Data
Testing set
38
© 2006 KDnuggets
Classification Step 2:
Build a model on a training
set THE PAST
Results Known
+
+ Training set
-
-
+
Data
Model Builder
Testing set
39
© 2006 KDnuggets
Classification Step 3:
Evaluate on test set (Re-
train?)
Results Known
+
+ Training set
-
-
+
Data
Model Builder
Evaluate
Predictions
+
Y N
-
+
Testing set -
40
© 2006 KDnuggets
Unbalanced data
Sometimes, classes have very unequal frequency
Attrition prediction: 97% stay, 3% attrite (in a month)
medical diagnosis: 90% healthy, 10% disease
eCommerce: 99% don’t buy, 1% buy
Security: >99.99% of Americans are not terrorists
41
© 2006 KDnuggets
Handling unbalanced data –
how?
If we have two classes that are very
unbalanced, then how can we
evaluate our classifier method?
42
© 2006 KDnuggets
Balancing unbalanced data,
1
With two classes, a good approach is to build
BALANCED train and test sets, and train
model on a balanced set
randomly select desired number of minority class
instances
add equal number of randomly selected majority
class
How do we generalize “balancing” to multiple
classes?
43
© 2006 KDnuggets
Balancing unbalanced data,
2
Generalize “balancing” to multiple
classes
Ensure that each class is represented with
approximately equal proportions in train
and test
44
© 2006 KDnuggets
A note on parameter tuning
It is important that the test data is not used in any
way to create the classifier
Some learning schemes operate in two stages:
Stage 1: builds the basic structure
Stage 2: optimizes parameter settings
45
© 2006 KDnuggets
Making the most of the data
Once evaluation is complete, all the data
can be used to build the final classifier
Generally, the larger the training data the
better the classifier (but returns diminish)
The larger the test data the more accurate
the error estimate
46
© 2006 KDnuggets
Classification:
Train, Validation, Test split
Results Known
+
Training set Model
+
-
-
Builder
+
Data
Evaluate
Model Builder
Predictions
+
-
Y N +
Validation set -
+
- Final Evaluation
+
Final Test Set Final Model -
47
© 2006 KDnuggets
Cross-validation
Cross-validation avoids overlapping test sets
First step: data is split into k subsets of equal size
Second step: each subset in turn is used for testing and
the remainder for training
48
© 2006 KDnuggets
Cross-validation example:
—Break up data into groups of the same size
—
—
—Hold aside one group for testing and use the rest to build
model
Test
—
—Repeat
49
49
© 2006 KDnuggets
More on cross-validation
Standard method for evaluation: stratified ten-fold
cross-validation
Why ten? Extensive experiments have shown that
this is the best choice to get an accurate estimate
Stratification reduces the estimate’s variance
Even better: repeated stratified cross-validation
E.g. ten-fold cross-validation is repeated ten times and
results are averaged (reduces the variance)
50
© 2006 KDnuggets
Direct Marketing Paradigm
Find most likely prospects to contact
Not everybody needs to be contacted
Number of targets is usually much smaller than
number of prospects
Typical Applications
retailers, catalogues, direct mail (and e-mail)
customer acquisition, cross-sell, attrition prediction
...
51
© 2006 KDnuggets
Direct Marketing Evaluation
Accuracy on the entire dataset is not
the right measure
Approach
develop a target model
score all prospects and rank them by decreasing
score
select top P% of prospects for action
Cumulative % Hits
90
Definition: 80
CPH(P,M) 70
= % of all targets 60
Random
50
in the first P% 40
of the list scored 30
20
by model M 10
CPH frequently 0
called Gains
5
45
65
95
15
25
35
55
75
85
Pct list
5% of random list have 5% of targets
54
© 2006 KDnuggets
CPH: Random List vs
Model-ranked list
100
Cumulative % Hits
90
80
70
60 Random
50 Model
40
30
20
10
0
5
45
65
95
15
25
35
55
75
85
Pct list
5% of random list have 5% of targets,
but 5% of model ranked list have 21% of targets
CPH(5%,model)=21%.
55
© 2006 KDnuggets
Lift
Lift(P,M) = CPH(P,M) / P
15
25
65
35
45
55
75
85
95
we call CPH.
P -- percent of the list
56
© 2006 KDnuggets
Lift – a measure of model
quality
Lift helps us decide which models are better
If cost/benefit values are not available or
changing, we can use Lift to select a better
model.
Model with the higher Lift curve will
generally be better
57
© 2006 KDnuggets
Clustering
© 2006 KDnuggets
Clustering
Unsupervised learning:
Finds “natural” grouping of
instances given un-labeled data
59
© 2006 KDnuggets
Clustering Methods
Many different method and algorithms:
For numeric and/or symbolic data
Deterministic vs. probabilistic
Exclusive vs. overlapping
Hierarchical vs. flat
Top-down vs. bottom-up
60
© 2006 KDnuggets
Clustering Evaluation
Manual inspection
Benchmarking on existing labels
Cluster quality measures
distance measures
high similarity within a cluster, low across
clusters
61
© 2006 KDnuggets
The distance function
Simplest case: one numeric attribute A
Distance(X,Y) = A(X) – A(Y)
62
© 2006 KDnuggets
Simple Clustering: K-means
Works with numeric data only
1) Pick a number (K) of cluster centers (at
random)
2) Assign every item to its nearest cluster
center (e.g. using Euclidean distance)
3) Move each cluster center to the mean of
its assigned items
4) Repeat steps 2,3 until convergence
(change in cluster assignments less than a
threshold)
63
© 2006 KDnuggets
K-means example, step 1
1
c1
Y
Pick 3 c2
initial
cluster
centers
(randomly)
c3
X
64
© 2006 KDnuggets
K-means example, step 2
c1
Y
c2
Assign
each point
to the closest
cluster
center c3
X
65
© 2006 KDnuggets
K-means example, step 3
c1 c1
Y
Move c2
each cluster
center c3
c2
to the mean
of each cluster c3
X
66
© 2006 KDnuggets
K-means example, step 4a
Reassign c1
points Y
closest to a
different new
cluster center
c3
Q: Which c2
points are
reassigned?
X
67
© 2006 KDnuggets
K-means example, step 4b
Reassign c1
points Y
closest to a
different new
cluster center
c3
Q: Which c2
points are
reassigned?
X
68
© 2006 KDnuggets
K-means example, step 4c
1
c1
Y
3
A: three
points with 2
animation c3
c2
X
69
© 2006 KDnuggets
K-means example, step 4d
c1
Y
re-compute
cluster
means c3
c2
X
70
© 2006 KDnuggets
K-means example, step 5
c1
Y
c2
move cluster
centers to c3
cluster means
X
71
© 2006 KDnuggets
Data Mining
Applications
© 2006 KDnuggets
Problems Suitable for Data-
Mining
require knowledge-based decisions
have a changing environment
have sub-optimal current methods
have accessible, sufficient, and relevant
data
provides high payoff for the right
decisions!
73
© 2006 KDnuggets
Major Application Areas for
Data Mining Solutions
Advertising
Bioinformatics
Customer Relationship Management (CRM)
Database Marketing
Fraud Detection
eCommerce
Health Care
Investment/Securities
Manufacturing, Process Control
Sports and Entertainment
Telecommunications
Web
74
© 2006 KDnuggets
Application: Search Engines
Before Google, web search engines used mainly
keywords on a page – results were easily subject
to manipulation
Google's early success was partly due to its
algorithm which uses mainly links to the page
Google founders Sergey Brin and Larry Page
were students at Stanford in 1990s
Their research in databases and data mining led
to Google
75
© 2006 KDnuggets
Microarrays: Classifying
Leukemia
Leukemia: Acute Lymphoblastic (ALL) vs
Acute Myeloid (AML), Golub et al, Science,
v.286, 1999
72 examples (38 train, 34 test), about 7,000
genes
ALL AML
77
© 2006 KDnuggets
Application:
Direct Marketing and
CRM
Most major direct marketing companies are using
modeling and data mining
Most financial companies are using customer
modeling
Modeling is easier than changing customer
behaviour
Example
Verizon Wireless reduced customer attrition rate from
2% to 1.5%, saving many millions of $
78
© 2006 KDnuggets
Application: e-Commerce
Amazon.com recommendations
if you bought (viewed) X, you are likely to buy Y
Netflix
If you liked "Monty Python and the Holy Grail",
79
© 2006 KDnuggets
Application:
Security and Fraud
Detection
Credit Card Fraud Detection
over 20 Million credit cards protected by
Neural networks (Fair, Isaac)
80
© 2006 KDnuggets
Data Mining, Privacy, and
Security
TIA: Terrorism (formerly Total) Information
Awareness Program –
TIA program closed by Congress in 2003
because of privacy concerns
However, in 2006 we learn that NSA is
analyzing US domestic call info to find
potential terrorists
Invasion of Privacy or Needed Intelligence?
81
© 2006 KDnuggets
Criticism of Analytic
Approaches to Threat
Detection:
Data Mining will
be ineffective - generate millions of false
positives
and invade privacy
82
© 2006 KDnuggets
Can Data Mining and
Statistics be Effective for
Threat Detection?
Criticism: Databases have 5% errors, so
analyzing 100 million suspects will generate 5
million false positives
Reality: Analytical models correlate many
items of information to reduce false positives.
Example: Identify one biased coin from 1,000.
After one throw of each coin, we cannot
After 30 throws, one biased coin will stand out with
high probability.
Can identify 19 biased coins out of 100 million with
sufficient number of throws
83
© 2006 KDnuggets
Another Approach: Link
Analysis
85
© 2006 KDnuggets
Data Mining with Privacy
Data Mining looks for patterns, not people!
Technical solutions can limit privacy invasion
Replacing sensitive personal data with anon. ID
Give randomized outputs
Multi-party computation – distributed data
…
86
© 2006 KDnuggets
The Hype Curve for
Data Mining and Knowledge
Discovery
Over-inflated
expectations
Growing acceptance
and mainstreaming
rising
expectations
Performance
Expectations
Disappointment
1990
1998 2000 2002
2005
87
© 2006 KDnuggets
Summary
Data Mining and Knowledge Discovery are
needed to deal with the flood of data
Knowledge Discovery is a process !
Avoid overfitting (finding random patterns
by searching too many possibilities)
88
© 2006 KDnuggets
Additional Resources
www.KDnuggets.com
data mining software, jobs, courses, etc
www.acm.org/sigkdd
ACM SIGKDD – the professional society for
data mining
89
© 2006 KDnuggets